今天在MySQL同步中再次遇到了令人讨厌的 1062与1032错误。。

对于MySQL的replication，感觉有点不太靠谱，，我不知道其他DBA都是用哪些同步方案。

在思考是否需要换个同步方案，随着访问量的增加，一主一备看来也比较脆弱了。宕机风险也高。

先前态度比较乐观。根据错误的提示：

20131128_14:56:09mysql> show slave status\G;
20131128_14:56:09*************************** 1. row ***************************
20131128_14:56:09               Slave_IO_State: Waiting for master to send event
20131128_14:56:09                  Master_Host: 192.168.101.210
20131128_14:56:09                  Master_User: backup
20131128_14:56:09                  Master_Port: 3306
20131128_14:56:09                Connect_Retry: 60
20131128_14:56:09              Master_Log_File: mysql-bin.001472
20131128_14:56:09          Read_Master_Log_Pos: 339328924
20131128_14:56:09               Relay_Log_File: hostname-relay-bin.004513
20131128_14:56:09                Relay_Log_Pos: 66635985
20131128_14:56:09        Relay_Master_Log_File: mysql-bin.001472
20131128_14:56:09             Slave_IO_Running: Yes
20131128_14:56:09            Slave_SQL_Running: No
20131128_14:56:09              Replicate_Do_DB: 
20131128_14:56:09          Replicate_Ignore_DB: mysql,test
20131128_14:56:09           Replicate_Do_Table: 
20131128_14:56:09       Replicate_Ignore_Table: 
20131128_14:56:09      Replicate_Wild_Do_Table: 
20131128_14:56:09  Replicate_Wild_Ignore_Table: 
20131128_14:56:09                   Last_Errno: 1032
20131128_14:56:09                   Last_Error: Could not execute Update_rows event on table feiliu_bbs.uc_member_info; Duplicate entry '20551928' fo20131128_14:56:09r key 'PRIMARY', Error_code: 1062; Can't find record in 'uc_member_info', Error_code: 1032; handler error HA_ERR_END_OF_FILE; the ev20131128_14:56:09ent's master log mysql-bin.001472, end_log_pos 66636646
20131128_14:56:09                 Skip_Counter: 0
20131128_14:56:09          Exec_Master_Log_Pos: 66635839
20131128_14:56:09              Relay_Log_Space: 339336281
20131128_14:56:09              Until_Condition: None
20131128_14:56:09               Until_Log_File: 
20131128_14:56:09                Until_Log_Pos: 0
20131128_14:56:09           Master_SSL_Allowed: No
20131128_14:56:09           Master_SSL_CA_File: 
20131128_14:56:09           Master_SSL_CA_Path: 
20131128_14:56:09              Master_SSL_Cert: 
20131128_14:56:09            Master_SSL_Cipher: 
20131128_14:56:09               Master_SSL_Key: 
20131128_14:56:09        Seconds_Behind_Master: NULL
20131128_14:56:09Master_SSL_Verify_Server_Cert: No
20131128_14:56:09                Last_IO_Errno: 0
20131128_14:56:09                Last_IO_Error: 
20131128_14:56:09               Last_SQL_Errno: 1032
20131128_14:56:09               Last_SQL_Error: Could not execute Update_rows event on table feiliu_bbs.uc_member_info; Duplicate entry '20551928' fo20131128_14:56:09r key 'PRIMARY', Error_code: 1062; Can't find record in 'uc_member_info', Error_code: 1032; handler error HA_ERR_END_OF_FILE; the ev20131128_14:56:09ent's master log mysql-bin.001472, end_log_pos 66636646
20131128_14:56:09  Replicate_Ignore_Server_Ids: 
20131128_14:56:09             Master_Server_Id: 1

发现时主键重复与更新失败。

主键重复的情况按照常理说可以直接跳过，所以我写好了命令：

命令1：stop slave sql_thread;set global sql_slave_skip_counter=1;start slave sql_thread;

Mysql 的Replication主要两个线程：1:IO_Thread 2:SQL_Thread;

网上都是建议直接stop slave或者start slave.这里主要是sql_thread的异常中断，所以我只重启sql_thread;

然后执行过命令1之后，发现这种情况不停的发生。不停的一次次跳过太过繁琐，通过问题来看主要是针对表

feiliu_bbs.uc_member_info

最终实在忍受不了，就打算从主库dump一份最新的数据到备份。于是我做了如下操作：

0：停止slave的同步进程

stop slave;

1：在master端手动备份目标表：uc_member_info,改名为uc_member_info_bak

create table uc_member_info_bak

select * from uc_member_info where 1=2;

2：导出表

mysqldump -uroot -p dbname tablename > tablename.sql

3:scp到备库，然后导入

mysql>source tablename.sql;

4:然后切换表名

切换之前，在新表uc_member_info_bak中补加原表uc_member_info 中的索引；

补加表中原有的索引：

create index idx_xxxx on uc_member_info_bak(cloumn_name);

切换表名：

alter table uc_member_info rename to uc_member_info_old;

alter table uc_member_info_bak rename to uc_member_info;

5:开启同步进程

start slave;

由于my.cnf中已经设置跳过主键重复错误。

slave-skip-errors = 1062

error错误日志中会不停弹出如下错误信息。

130922 12:23:20 [Warning] Slave SQL: Could not execute Write_rows event on table feiliu_bbs.uc_member_info; Duplicate entry '1756491
4' for key 'PRIMARY', Error_code: 1062; handler error HA_ERR_FOUND_DUPP_KEY; the event's master log mysql-bin.001185, end_log_pos 69
4223737, Error_code: 1062
130922 12:23:21 [Warning] Slave SQL: Could not execute Write_rows event on table feiliu_bbs.uc_member_info; Duplicate entry '1756491
5' for key 'PRIMARY', Error_code: 1062; handler error HA_ERR_FOUND_DUPP_KEY; the event's master log mysql-bin.001185, end_log_pos 69
4274279, Error_code: 1062
130922 12:23:21 [Warning] Slave SQL: Could not execute Write_rows event on table feiliu_bbs.uc_member_info; Duplicate entry '1756491
6' for key 'PRIMARY', Error_code: 1062; handler error HA_ERR_FOUND_DUPP_KEY; the event's master log mysql-bin.001185, end_log_pos 69
4278383, Error_code: 1062
130922 12:23:21 [Warning] Slave SQL: Could not execute Write_rows event on table feiliu_bbs.uc_member_info; Duplicate entry '1756491
7' for key 'PRIMARY', Error_code: 1062; handler error HA_ERR_FOUND_DUPP_KEY; the event's master log mysql-bin.001185, end_log_pos 69
4326934, Error_code: 1062
130922 12:23:21 [Warning] Slave SQL: Could not execute Write_rows event on table feiliu_bbs.uc_member_info; Duplicate entry '1756491
8' for key 'PRIMARY', Error_code: 1062; handler error HA_ERR_FOUND_DUPP_KEY; the event's master log mysql-bin.001185, end_log_pos 69
4340338, Error_code: 1062

弹出这个问题的原因是

备库现有的的uc_member_info是从主库dump过来的，比原有的备库表uc_member_info_old 进度要快。因为备库与主库之间存在较长时间的间隔，主库的binlog，在同步停止的时间内，没有及时发到备库导致。

解决方法：

首先停止同步，然后根据最后一条错误信息：

130922 12:23:21 [Warning] Slave SQL: Could not execute Write_rows event on table feiliu_bbs.uc_member_info; Duplicate entry '1756491
8' for key 'PRIMARY', Error_code: 1062; handler error HA_ERR_FOUND_DUPP_KEY; the event's master log mysql-bin.001185, end_log_pos 69
4340338, Error_code: 1062

删除 主键>=“17564918“的数据。因为主键为自增长。但是这个自增长不依赖自动序列。而上来源于另外一张表的主键。索引不用考虑自增序列的混乱情况。

删除之后，恢复同步。就不在出现主键重复的错误提示了。

等待同步，大约半个小时之后，仍然会出现1032错误。仍旧是表uc_member_info.现在怀疑应该不单单是单表的问题，而是数据库本身的一致性已经处理问题。准备放弃，进行重做。但是目前不能让备库停止，不能影响其他数据库的正常读取业务。故准备采用杀手锏！

准备进行设置数据库跳过一般的错误异常，使之不会轻易停止同步。

使用：slave_exec_mode参数

（具体参数说明 http://blog.csdn.net/zhangbiaobiaobiao/article/details/17072199 这里记录过大概。）

然后准备后续的主备的一致性验证与备份重做。